An assistant for the lectures
Instead of leafing through hundreds of recordings by hand, you ask a question — and get ready-made fragments from the lectures on your topic.
For the last few weeks we have been building a new assistant for Śrīla Prabhupāda’s lectures inside the app. The idea is simple: instead of leafing through hundreds of recordings by hand, you just ask — and get ready-made fragments from the lectures on your topic.
What already works
Semantic search over the transcripts. Ask “what did Prabhupāda say about varṇāśrama” and the assistant finds the relevant fragments and quotes them, each marked with its lecture and timecode.
Every citation is a small player. You can listen to the fragment itself, open the full lecture from that exact moment, save the fragment to your notes, or add the whole lecture to a playlist.
Catalogue filters: author, place, date, source, tag. “Lectures in Bombay, 1973” — and the list is ready.
A table of contents for each lecture, by chapter and with timecodes — so you can see the shape of a recording at a glance and jump to the part you want.
Personal context: “where did I stop”, “what did I listen to this week”, “what should I hear next” — the assistant sees your listening history and makes suggestions.
It works in Russian and English, across the whole corpus of Śrīla Prabhupāda’s lectures.
Under the hood
Underneath is a vector search over chunks of transcript, plus an LLM agent with a
set of tools it chooses between as the question demands. The real toolset is
roughly: a hybrid semantic-plus-lexical search (chunks_search), a deterministic
lookup for scripture addresses like “BG 2.13” (chunks_get_by_address), a
chapter outline (track_outline_get), catalogue resolvers (author, source,
location, tag), and the personal-history tools (user_history_search,
user_tracks_list). Recommendations run as their own step in the graph.
What is not here yet: voice input and shareable conversations — both are planned.
We will show it in action soon.
Part of
Lectorium